What GPU do I need to run meta-llama/Llama-Guard-4-12B?
A 12B language model for chat and instruction-following. 12.0B parameters, published in BF16. View on Hugging FaceGated
Llama-Guard-4-12B is published by meta-llama on Hugging Face, with 46,640 downloads and 128 likes to date. It's a Llama4ForConditionalGeneration model built for image-text-to-text, published natively in BF16, and gated: you'll need to accept the model's terms on Hugging Face before downloading weights.
What Llama-Guard-4-12B is
Llama-Guard-4-12B is a 12B-parameter language model published by Meta on Hugging Face. It is a content-moderation/safety classifier rather than a general chat model. It is released under Custom license.
License note: a lab-specific license (tagged "other" on Hugging Face); read the model card's own license section before commercial use. Facts in this section are sourced from Llama-Guard-4-12B's Hugging Face model card, not benchmarked by Aquanode.
What it's used for
- Content moderation
- Safety classification for chat pipelines
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision times a flat 1.2 overhead for activations and fragmentation. The KV-cache grows with context and is not in that factor; it is listed per model below where the architecture is published. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| BF16 | 22.4 GB | 26.8 GB | RTX A6000 | 1 | $0.363/hr |
| cheaper alt. | RTX 3060 | 3 | $0.330/hr | ||
| FP8 (quantized) | 11.2 GB | 13.4 GB | RTX 4000 SFF Ada | 1 | $0.198/hr |
| INT4 (quantized) | 5.6 GB | 6.7 GB | RTX 3060 | 1 | $0.110/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Cheapest way to run Llama-Guard-4-12B at its published (BF16) precision: 1× RTX A6000, at $0.363/hr per GPU ($0.363/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
Llama-Guard-4-12B: KV cache by context length
The KV cache is the memory the attention layers hold for every token of context, on top of the weights. It is not part of the flat 1.2x overhead in the table above, grows with context length and with every concurrent request, and is why a long-context deployment needs more VRAM than the table shows.
Grouped-query attention: every layer caches keys and values for a few shared KV heads. Cached per token: 196,608 bytes at 16-bit.
| Context | KV cache, one sequence |
|---|---|
| 4K | 0.75 GB |
| 32K | 6.0 GB |
| 128K | 24.0 GB |
| 10M (model maximum) | 1920.0 GB |
Computed from the layer, head and window counts in the model's own configuration (unsloth/Llama-Guard-4-12B), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
Llama-Guard-4-12B: common questions
Does Llama-Guard-4-12B fit on a 32 GB GPU?
Yes. At BF16 it needs 26.8 GB of VRAM, so a 32 GB card holds it with 5.2 GB to spare. A 24 GB card is not enough for it at BF16.
Do I need approval to download Llama-Guard-4-12B?
Yes. meta-llama gates this repository on Hugging Face, so you have to accept its terms with a signed-in account before the weights will download. It does not change the 26.8 GB the model needs once you have them.
What is the least VRAM Llama-Guard-4-12B can run in?
6.7 GB, at INT4 (quantized), which fits an 8 GB card, against 26.8 GB at BF16. That row assumes an INT4 checkpoint has actually been published for this model, so check its Hugging Face page before planning around it.
Does quantizing Llama-Guard-4-12B lower the GPU bill?
Yes. At BF16 the cheapest live fit is one RTX A6000 at $0.363/hr. At INT4 (quantized) it drops to one RTX 3060 at $0.110/hr, provided a quantized checkpoint exists for it.
How to run Llama-Guard-4-12B
Run Llama-Guard-4-12B with vLLM
Generic example, not from the model's own docs: adjust flags (quantization, context length, parallelism) for your setup.
vllm serve meta-llama/Llama-Guard-4-12B --tensor-parallel-size 1Deploy Llama-Guard-4-12B on Aquanode
Aquanode has no one-click deploy template for Llama-Guard-4-12B; you install the inference engine yourself with the commands below. Aquanode sells GPU pods billed per second, not a hosted inference API.
- Launch a bare GPU pod sized to the requirement above (1× RTX A6000 or larger).
- Open a terminal on the pod, or save one of the commands above as a startup script so it runs automatically the first time the pod boots.
- Run the command and connect to the resulting endpoint.
Launch a GPU podWeight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Llama Guard models
- Llama-Guard-3-8B (8.0B, BF16)
- Llama-Guard-3-1B (1.5B, BF16)
- Llama-Phishsense-1B (1.2B, BF16)
Alternatives at this size
Other models for image-text-to-text within about a third of Llama-Guard-4-12B's 12.0B parameters, from other model lines.
- Qwen3.5-9B (9.7B, BF16)
- gemma-3-12b-it (12.2B, BF16)
- UI-TARS-1.5-7B (8.3B, F32)
- Cosmos-Reason2-8B (8.8B, BF16)
- Ovis2.5-9B (9.2B, BF16)
More on Llama-Guard-4-12B
Fits on a 32 GB GPU at BF16: every model that fits in 32 GB.
Fits on a 16 GB GPU at FP8: every model that fits in 16 GB.
Fits on an 8 GB GPU at INT4: every model that fits in 8 GB.
Best vision-language models: how Llama-Guard-4-12B ranks against the rest.
Related reading: RTX A6000 pricing and specs, How much VRAM you need for LLMs, Serving LLMs with vLLM, vLLM vs TensorRT-LLM vs SGLang, and Best GPU for LLM inference.
